Embed Size (px)
Citation preview
- 1 -CS - ES
Embedded Systems
- 2 -CS - ES
Message Sequence Charts
REVIEW
- 3 -CS - ES
Message Sequence Charts
Message Sequence Charts (MSC) is a language to describe the interaction between a number of independent message-passing instances.
Defined by ITU (International Telecommunication Union) - Z.120 recommendation
MSC is a scenario language graphical formal practical widely applicable
REVIEW
- 4 -CS - ES
MSC
In telecommunication industry, MSCs are the first choice to describe example traces of the system under development. MSCs are used throughout the whole protocol life cycle from requirements analysis to testing.
To define longer traces hierarchically, simple MSCs can be composed by operators in high-level MSC (HMSC).
Message Sequence Charts may be used for requirement specification, simulation and validation, test-case specification and documentation of real-time systems.
REVIEW
- 5 -CS - ES
Message sequence charts (MSC)
Graphical means for representing schedules; time used vertically, “geographical” distribution horizontally.
REVIEW
- 6 -CS - ES
Basic MSC in a nutshell
User AC System
Code
OK
msc User_accepted
UnlockCard out
Idle
Door unlocked
MSC diagram
MSC heading
Conditionno predicate logic,
merely a label
Output event
Input event
Instance
Message to the
environment
Instance end
REVIEW
- 7 -CS - ES
Timer set and timeout
• User is accepted forget to push the door
• AC system will detect this through the expiration of the timer Lock
REVIEW
- 8 -CS - ES
Preferred situation REVIEW
- 9 -CS - ES
MSC reference
• In almost all description/programming/specification languages there is a way to isolate subparts of the description in a separate named construct (procedures, functions, classes, packages)
• In MSC there are MSCs which can be referred from other MSCs.
REVIEW
- 10 -CS - ES
MSC reference
• Assume that the scenario where the user is accepted is part of a larger context where there is an automatic door. When the door is unlocked it automatically opens.
• The MSC reference symbol is a box with rounded corners.
REVIEW
- 11 -CS - ES
HMSC (High Level MSC)
User accepted
Idle
Unlocked_reset Unlocked_timeout
Door unlocked
Unlocked_unclosed
User rejected
msc ACsystemOverview
HMSC Start
MSC Reference
Condition
Alternative
Loop
Connection Point
REVIEW
- 12 -CS - ES
Data in MSC-2000
MSC has no data language of its own!
MSC has parameterized data languages such that
fragments of your favorite (data) language can be used• C, C++, SDL, Java, ...
MSC can be parsed without knowing the details of the chosen data language
the interface between MSC and the chosen data language is given in a set of interface functions
REVIEW
- 13 -CS - ES
Data Flow Models
REVIEW
- 14 -CS - ES
Data flow modeling
Def.: The process of identifying, modeling and documenting how data moves around an information system.
Data flow modeling examines processes (activities that transform data from one form to
another), data stores (the holding areas for data), external entities (what sends data into a system or receives data
from a system, and data flows (routes by which data can flow).
REVIEW
- 15 -CS - ES
Dataflow model Nodes represent transformations
May execute concurrently
Edges represent flow of tokens (data) from one node to another May or may not have token at any given time
When all of node’s input edges have at least one token, node may fire
When node fires, it consumes input tokens processes transformation and generates output token
Nodes may fire simultaneously
Several commercial tools support graphical languages for capture of dataflow model Can automatically translate to concurrent process model for
implementation Each node becomes a process
modulate convolve
transform
A B C D
Z
Nodes with more complex transformations
t1 t2
+ –
*
A B C D
Z
Nodes with arithmetic transformations
t1 t2
Z = (A + B) * (C - D)
REVIEW
- 16 -CS - ES
Philosophy of Dataflow Languages
Drastically different way of looking at computation
Von Neumann imperative language style: program counter controls everything
Dataflow language: movement of data the priority
Scheduling responsibility of the system, not the programmer
REVIEW
- 17 -CS - ES
Applications of Dataflow
signal-processing applications
Anything that deals with a continuous stream of data
Becomes easy to parallelize
Buffers typically used for signal processing applications anyway
REVIEW
- 18 -CS - ES
Kahn Process Networks
Proposed by Kahn in 1974 as a general-purpose scheme for parallel programming
Theoretical foundation for dataflow Unique attribute: deterministic
…Send();
…
…Wait();
…
REVIEW
- 19 -CS - ES
Properties of Kahn process networks (2)
There is only one sender per channel. A process cannot check whether data is available before
attempting a read. A process cannot wait for data for more than one port at a time. Therefore, the order of reads depends only on data, not on the
arrival time. Therefore, Kahn process networks are deterministic (!); for a
given input, the result will always the same, regardless of the speed of the nodes.
This is the key beauty of KPNs!
REVIEW
- 20 -CS - ES
Kahn Process Networks
Key idea:
Reading an empty channel blocks until data is available
No other mechanism for sampling communication channel’s contents
Can’t check to see whether buffer is empty Can’t wait on multiple channels at once
REVIEW
- 21 -CS - ES
Sample parallel program S
(1) … channel declation
processes f, g, h are declared
REVIEW
- 22 -CS - ES
A Kahn Process From Kahn’s original 1974 paper
process f(in int u, in int v, out int w){int i; bool b = true;for (;;) {i = b ? wait(u) : wait(v);printf("%i\n", i);send(i, w);b = !b;
}}
f
u
v
w
Process alternately reads from u and v, prints the data
value, and writes it to w
What does this do?
REVIEW
- 23 -CS - ES
A Kahn Process
From Kahn’s original 1974 paper:
process f(in int u, in int v, out int w){int i; bool b = true;for (;;) {i = b ? wait(u) : wait(v);printf("%i\n", i);send(i, w);b = !b;
}}
Process interface
includes FIFOs
wait() returns the next token in an input FIFO,
blocking if it’s empty
send() writes a data value on an output FIFO
REVIEW
- 24 -CS - ES
A Kahn Process
From Kahn’s original 1974 paper:
process g(in int u, out int v, out int w){int i; bool b = true;for(;;) {i = wait(u);if (b) send(i, v); else send(i, w);b = !b;
}}
guv
w
Process reads from u and alternately copies it to v and w
What does this do?
REVIEW
- 25 -CS - ES
A Kahn Process
From Kahn’s original 1974 paper:
process h(in int u, out int v, int init){int i = init;send(i, v);for(;;) {i = wait(u);send(i, v);
}}
hu v
Process sends initial value, then passes through values.
What does this do?
REVIEW
- 26 -CS - ES
Sample parallel program S
(1) … channel declation
processes f, g, h are declared
(6) … body of the main program: - calling instances of the
processes- actual names of the channels
are bound to the formal parameters- infix operator par concurrent
activation of the processes
REVIEW
- 27 -CS - ES
A Kahn System
What does this do?
fg
hinit = 0
hinit = 1
Emits a 1 then copies input to output
Emits a 0 then copies input to output
Prints an alternating sequence of 0’s and 1’s
T2 Z
T2
X
Y
REVIEW
- 28 -CS - ES
REVIEW
- 29 -CS - ES
Determinism
Process: “continuous mapping” of input sequence to output sequences
Continuity: process uses prefix of input sequences to produce prefix of output sequences. Adding more tokens does not change the tokens already produced
The state of each process depends on token values rather than their arrival time
Unbounded FIFO: the speed of the two processes does not affect the sequence of data values
Fx1,x2,x3… y1,y2,y3…
REVIEW
- 30 -CS - ES
Synchronous Dataflow (SDF) Edward Lee and David Messerchmitt, Berkeley, 1987
Ptolemy System
Restriction of Kahn Networks to allow compile-time scheduling
Basic idea: each process reads and writes a fixed number of tokens each time it fires:
loopread 3 A, 5 B, 1 C …compute…write 2 D, 1 E, 7 F
end loop
REVIEW
- 31 -CS - ES 31
Synchronous dataflow
With digital signal-processors (DSPs), data flows at fixed rate
ADC DACDSP
0110.. 1110..
REVIEW
- 32 -CS - ES 32
Synchronous dataflow Multiple tokens consumed and produced per firing
Synchronous dataflow model takes advantage of this Each edge labeled with number of tokens
consumed/produced each firing Can statically schedule nodes, so can easily use sequential
program model• Don’t need real-time operating system and its overhead
Algorithms developed for scheduling nodes into “single-appearance” schedules Only one statement needed to call each node’s associated
procedure• Allows procedure inlining without code explosion, thus reducing
overhead even more
modulate convolve
transform
A B C D
Z
Synchronous dataflow
mt1 ct2
mA mB mC mD
tZ
tt1 tt2
t1 t2
REVIEW
- 33 -CS - ES
SDF and Signal Processing
Restriction natural for multirate signal processing
Typical signal-processing processes:
Unit-rate• Adders, multipliers
Upsamplers (1 in, n out) Downsamplers (n in, 1 out)
- 34 -CS - ES
Asynchronous message passing:Synchronous data flow (SDF) Asynchronous message passing=
tasks do not have to wait until output is accepted. Synchronous data flow =
all tokens are consumed at the same time.
SDF model allows static scheduling of token production and consumption.
In the general case, buffers may be needed at edges.
- 35 -CS - ES
Synchronous DataFlow
Actor enabling = each incoming arc carries at least weight tokens
Actor execution = atomic consumption/production of tokens by an enabled actor i.e., consume weight tokens on each incoming arcs and
produce weight tokens on each outgoing arc Delay is an initial token load on an arc.
SDF firing rules:
- 36 -CS - ES
SDF Example
AABAACC
BStatic schedule:
- 37 -CS - ES
Parallel Scheduling of SDF Models
A
C
D
B
Sequentialperiodic admissible sequential
schedule (PASS)
Parallelperiodic admissible parallel
schedule (PAPS)
SDF is suitable for automated mapping onto
parallel processors and
synthesis of parallel circuits.
(admissible = correct schedule, finite amount of memory required)
- 38 -CS - ES
SDF Scheduling AlgorithmLee/Messerschmitt 1987
1. Establish relative execution rates Generate balance equations Solve for smallest positive integer vector q
2. Determine periodic schedule Form an arbitrarily ordered list of all nodes in the system Repeat:
• For each node in the list, schedule it if it is runnable, trying each node once
• If each node has been scheduled qn times, stop.• If no node can be scheduled, indicate deadlock.
Source: Lee/Messerschmitt, Synchronous Data Flow (1987)
- 39 -CS - ES
Multi-rate SDF System
DAT (digital audio tape) -to-CD rate converter Converts a 44.1 kHz sampling rate to 48 kHz
1 1 2 3 2 7 8 7 5 1
Upsampler Downsampler
- 40 -CS - ES
SDF: restriction of Kahn networks
An SDF graph is a tuple (V, E, cons, prod, d) where V is a set of nodes (activities) E is a set of edges (buffers) cons: E N number of tokens consumed prod: E N number of tokens produced d: E N number of initial tokens
d: „delay“ (sample offset between input and output)
- 41 -CS - ES
Delays
Kahn processes often have an initialization phase
SDF doesn’t allow this because rates are not always constant
Alternative: an SDF system may start with tokens in its buffers
These behave like delays (signal-processing)
Delays are sometimes necessary to avoid deadlock
- 42 -CS - ES
Example SDF System Finite Impulse Response
FIR Filter (all single-rate)
dup
*c0
dup
*c1
+
dup
*c2
+
dup
*c3
+
*c(N-1)
+
One-cycle delayDuplicate
Constant multiply
(filter coefficient)
Adder
yn
xn
Yn = xn*c0 + xn-1*c1 + … + xn-(N-1)*c(N-1)
…
…
…
- 43 -CS - ES
SDF Scheduling
Schedule can be determined completely before the system runs
Two steps:
1. Establish relative execution rates by solving a system of linear equations
2. Determine periodic schedule by simulating system for a single round
- 44 -CS - ES
SDF Scheduling
Goal: a sequence of process firings that: Runs each process at least once in proportion to its rate Avoids underflow
• no process fired unless all tokens it consumes are available Returns the number of tokens in each buffer to their initial state
Result: the schedule can be executed repeatedly without accumulating tokens in buffers
- 45 -CS - ES
Balance equations
Number of produced tokens must equal number of consumed tokens on every edge
Repetitions (or firing) vector vS of schedule S: number of firings of each actor in S vS(A) np = vS(B) nc
must be satisfied for each edge
np ncA B
- 46 -CS - ES
Balance equations
B C
A3
1
1
1
22
11
Balance for each edge: 3 vS(A) - vS(B) = 0 vS(B) - vS(C) = 0 2 vS(A) - vS(C) = 0 2 vS(A) - vS(C) = 0
- 47 -CS - ES
Balance equations
M vS = 0iff S is periodic Full rank (as in this case)
• no non-zero solution • no periodic schedule
(too many tokens accumulate on A->B or B->C)
3 -1 00 1 -12 0 -12 0 -1
M =
B C
A3
1
1
1
22
11
topology matrix
the (c, r)th entry in the matrix is the amount of data produced by node c on arc
r each time it is involved
- 48 -CS - ES
Balance equations
Non-full rank• infinite solutions exist
Any multiple of vS = |1 2 2|T satisfies the balance equations ABCBC and ABBCC are minimal valid schedules
2 -1 00 1 -12 0 -12 0 -1
M =
B C
A2
1
1
1
22
11
- 49 -CS - ES
Static SDF scheduling
Main SDF scheduling theorem (Lee ‘86): A connected SDF graph with n actors has a
periodic schedule iff its topology matrix M has rank n-1 If M has rank n-1 then there exists a unique
smallest integer solution vS to M vS = 0
Rank must be at least n-1 because we need at least n-1 edges (connected-ness), providing each a linearly independent row Admissibility is not guaranteed, and depends
on initial tokens on cycles
- 50 -CS - ES
Admissibility of schedules
No admissible schedule:BACBA, then deadlock… Adding one token on A->C makes
BACBACBA valid Making a periodic schedule admissible is always
possible, but changes specification...
B C
A1
2
1
3
2
3
- 51 -CS - ES
An Inconsistent System
No way to execute it without an unbounded accumulation of tokens
Only consistent solution is “do nothing”
b
1
ca1
32
1
1
a – c = 0a – 2b = 03b – c = 0
3a – 2c = 0
- 52 -CS - ES
Calculating Rates
Each arc imposes a constraint
b
d
12
3
2
c
a
3
41
3
21
6
3a – 2b = 04b – 3d = 0
b – 3c = 02c – a = 0d – 2a = 0
Solution:a = 2cb = 3cd = 4c
- 53 -CS - ES
Scheduling Example
Theorem guarantees any valid simulation will produce a schedule
b
d
12
3
2
c
a
3
41
3
21
6
a=2 b=3 c=1 d=4
Possible schedules:BBBCDDDDAABDBDBCADDABBDDBDDCAA… many more
BC … is not valid
- 54 -CS - ES
SDF Compiler
Task for an SDF compiler: Allocation of memory for the passing of data between nodes Scheduling of nodes onto processors in such a way that data is
available for a block when it is invoked
Assumptions on the SDF graph: The SDF graph is nonterminating and does not deadlock The SDF graph is connected
Goal: Development of a periodic admissible parallel schedule (PAPS) or a periodic admissible sequential schedule (PASS)
(admissible = correct schedule, finite amount of memory required)
- 55 -CS - ES
Does a PASS exist?
- 56 -CS - ES
Does a PASS exist?
A PSS (periodic sequential schedule) exists if rank (M) = s -1, where s is the number if nodes in a graph.
There is a v such that Mv = O where O is a vector full of zeros. v describes the number of firings in each scheduling period.
- 57 -CS - ES
A PASS exists
The rank of the matrix M = is s – 1 = 2 and v =
A valid schedule is Φ = {a, b, c, c}, but not Φ = {b, a, c, c}
The maximum buffer sizes for the arcs are b =< 1, 2, 2 >
- 58 -CS - ES
A PASS does not exist
The graph has sample rate inconsistencies.
A schedule for the graph will result in unbounded buffer sizes.
No PASS can be found (rank (M) = s = 3).
- 59 -CS - ES
PAPS
Assumption: Block 1 : 1 time unitBlock 2 : 2 time unitsBlock 3 : 3 time units
Trivial Case - All computations are scheduled on same processor
- 60 -CS - ES
PAPS
The performance can be improved, if a schedule is constructed that exploits the potential parallelism in the SDF-graph. Here the schedule covers one single period.
Single Period Schedule
- 61 -CS - ES
PAPS
The performance can be further improved, if the schedule is constructed over two periods.
Double Period Schedule
